diff options
| author | Raphael M Zinsly <rzinsly@linux.ibm.com> | 2020-11-12 13:12:24 -0300 |
|---|---|---|
| committer | Tulio Magno Quites Machado Filho <tuliom@linux.ibm.com> | 2020-11-12 13:16:36 -0300 |
| commit | 7beee7b39adeda657f45989b0635033dae25a1fd (patch) | |
| tree | 003641c1011e226ea6e5234c560f302e6d3f00ff /sysdeps/powerpc/powerpc64/le/power9/strncpy.S | |
| parent | b9d83bf3eb57e1cf8ef785f1a58e13ddf162b6f3 (diff) | |
| download | glibc-7beee7b39adeda657f45989b0635033dae25a1fd.tar.gz glibc-7beee7b39adeda657f45989b0635033dae25a1fd.tar.xz glibc-7beee7b39adeda657f45989b0635033dae25a1fd.zip | |
powerpc: Add optimized stpncpy for POWER9
Add stpncpy support into the POWER9 strncpy. Reviewed-by: Matheus Castanho <msc@linux.ibm.com> Reviewed-by: Tulio Magno Quites Machado Filho <tuliom@linux.ibm.com>
Diffstat (limited to 'sysdeps/powerpc/powerpc64/le/power9/strncpy.S')
| -rw-r--r-- | sysdeps/powerpc/powerpc64/le/power9/strncpy.S | 68 |
1 files changed, 67 insertions, 1 deletions
diff --git a/sysdeps/powerpc/powerpc64/le/power9/strncpy.S b/sysdeps/powerpc/powerpc64/le/power9/strncpy.S index cbfc37bda3..b4ba428662 100644 --- a/sysdeps/powerpc/powerpc64/le/power9/strncpy.S +++ b/sysdeps/powerpc/powerpc64/le/power9/strncpy.S @@ -18,11 +18,19 @@ #include <sysdep.h> +#ifdef USE_AS_STPNCPY +# ifndef STPNCPY +# define FUNC_NAME __stpncpy +# else +# define FUNC_NAME STPNCPY +# endif +#else # ifndef STRNCPY # define FUNC_NAME strncpy # else # define FUNC_NAME STRNCPY # endif +#endif /* !USE_AS_STPNCPY */ #ifndef MEMSET /* For builds without IFUNC support, local calls should be made to internal @@ -41,6 +49,12 @@ char * [r3] strncpy (char *dest [r3], const char *src [r4], size_t n [r5]) + or + + char * [r3] stpncpy (char *dest [r3], const char *src [r4], size_t n [r5]) + + if USE_AS_STPNCPY is defined. + The implementation can load bytes past a null terminator, but only up to the next 16-byte aligned address, so it never crosses a page. */ @@ -66,7 +80,15 @@ ENTRY (FUNC_NAME, 4) /* Empty/1-byte string optimization */ cmpdi r5,0 +#ifdef USE_AS_STPNCPY + bgt L(cont) + /* Compute pointer to last byte copied into dest. */ + addi r3,r3,1 + blr +L(cont): +#else beqlr +#endif addi r4,r4,1 neg r7,r4 @@ -96,12 +118,20 @@ ENTRY (FUNC_NAME, 4) sldi r10,r5,56 /* stxvl wants size in top 8 bits */ stxvl 32+v0,r11,r10 /* Partial store */ +#ifdef USE_AS_STPNCPY + /* Compute pointer to last byte copied into dest. */ + add r3,r11,r5 +#endif blr L(null): sldi r10,r8,56 /* stxvl wants size in top 8 bits */ stxvl 32+v0,r11,r10 /* Partial store */ +#ifdef USE_AS_STPNCPY + /* Compute pointer to last byte copied into dest. */ + add r3,r11,r7 +#endif add r11,r11,r8 sub r5,r5,r8 b L(zero_padding) @@ -185,6 +215,10 @@ L(n_tail4): sldi r10,r5,56 /* stxvl wants size in top 8 bits */ addi r11,r11,48 /* Offset */ stxvl 32+v3,r11,r10 /* Partial store */ +#ifdef USE_AS_STPNCPY + /* Compute pointer to last byte copied into dest. */ + add r3,r11,r5 +#endif blr L(prep_n_tail1): @@ -196,6 +230,10 @@ L(prep_n_tail1): L(n_tail1): sldi r10,r5,56 /* stxvl wants size in top 8 bits */ stxvl 32+v0,r11,r10 /* Partial store */ +#ifdef USE_AS_STPNCPY + /* Compute pointer to last byte copied into dest. */ + add r3,r11,r5 +#endif blr L(prep_n_tail2): @@ -209,6 +247,10 @@ L(n_tail2): sldi r10,r5,56 /* stxvl wants size in top 8 bits */ addi r11,r11,16 /* offset */ stxvl 32+v1,r11,r10 /* Partial store */ +#ifdef USE_AS_STPNCPY + /* Compute pointer to last byte copied into dest. */ + add r3,r11,r5 +#endif blr L(prep_n_tail3): @@ -223,6 +265,10 @@ L(n_tail3): sldi r10,r5,56 /* stxvl wants size in top 8 bits */ addi r11,r11,32 /* Offset */ stxvl 32+v2,r11,r10 /* Partial store */ +#ifdef USE_AS_STPNCPY + /* Compute pointer to last byte copied into dest. */ + add r3,r11,r5 +#endif blr L(prep_tail1): @@ -232,6 +278,10 @@ L(tail1): addi r9,r8,1 /* Add null terminator */ sldi r10,r9,56 /* stxvl wants size in top 8 bits */ stxvl 32+v0,r11,r10 /* Partial store */ +#ifdef USE_AS_STPNCPY + /* Compute pointer to last byte copied into dest. */ + add r3,r11,r8 +#endif add r11,r11,r9 sub r5,r5,r9 b L(zero_padding) @@ -246,6 +296,10 @@ L(tail2): sldi r10,r9,56 /* stxvl wants size in top 8 bits */ addi r11,r11,16 /* offset */ stxvl 32+v1,r11,r10 /* Partial store */ +#ifdef USE_AS_STPNCPY + /* Compute pointer to last byte copied into dest. */ + add r3,r11,r8 +#endif add r11,r11,r9 sub r5,r5,r9 b L(zero_padding) @@ -261,6 +315,10 @@ L(tail3): sldi r10,r9,56 /* stxvl wants size in top 8 bits */ addi r11,r11,32 /* offset */ stxvl 32+v2,r11,r10 /* Partial store */ +#ifdef USE_AS_STPNCPY + /* Compute pointer to last byte copied into dest. */ + add r3,r11,r8 +#endif add r11,r11,r9 sub r5,r5,r9 b L(zero_padding) @@ -276,6 +334,10 @@ L(tail4): sldi r10,r9,56 /* stxvl wants size in top 8 bits */ addi r11,r11,48 /* offset */ stxvl 32+v3,r11,r10 /* Partial store */ +#ifdef USE_AS_STPNCPY + /* Compute pointer to last byte copied into dest. */ + add r3,r11,r8 +#endif add r11,r11,r9 sub r5,r5,r9 @@ -331,7 +393,8 @@ L(zero_padding_memset): ld r0,FRAMESIZE+16(r1) mr r3,r30 /* Restore the return value of strncpy, i.e.: - dest. */ + dest. For stpncpy, the return value is the + same as return value of memset. */ ld r30,FRAMESIZE-8(r1) /* Restore r30. */ /* Restore the stack frame. */ addi r1,r1,FRAMESIZE @@ -342,3 +405,6 @@ L(zero_padding_memset): blr END (FUNC_NAME) +#ifndef USE_AS_STPNCPY +libc_hidden_builtin_def (strncpy) +#endif |